Monitoring and predicting important events in honeybee hives using acoustic signals and machine learning

Ruvinga, Stenford, Hunter, Gordon, Nebel, Jean-Christophe, Duran, Olga and Busquets, Rosa (2022) Monitoring and predicting important events in honeybee hives using acoustic signals and machine learning. In: UKAN+ Monitoring UK Biodiversity Symposium; 15 - 16 Jun 2022, Manchester, U.K.. (Unpublished)


Honeybees, as key pollinators, are of vital importance to both agriculture and the wider environment. According to the United Nations “Nearly 90% of the world’s wild flowering plant species depend, entirely, or at least in part, on animal pollination, along with more than 75% of the world’s food crops and 35% of global agricultural land.” The importance of honeybees extends beyond pollination to commercial pollination services and bee products such as honey, beeswax, pheromones, propolis and royal jelly. Over recent years, their populations have been in serious decline due to various factors, including pesticides, parasites, pollution, disease and possibly climate change and radio frequency or microwave electromagnetic radiation. Furthermore, traditional bee monitoring methods are invasive and can be stressful and even sometimes harmful to the bees. The queen bee is the lifeline of a honeybee colony as she is the only female who can lay fertilized eggs. If left queenless for any significant length of time, the colony will die off. The death or other loss of a healthy queen is thus very dangerous to the survival of a colony. Another major event in the natural lifecycle of a honeybee colony is the occurrence of a swarm. These normally occur shortly after a new queen has hatched, and usually result in a large number of worker bees leaving the hive along with the old queen. These are not popular with beekeepers, since they will tend to lose a substantial number of bees when swarms occur. Hence, successfully predicting that a swarm is going to occur soon would be of great interest to beekeepers. In this work, we contribute to addressing these problems by employing machine learning, statistical and signal processing approaches applied to audio data recorded from “queen-absent” and “queen-present” hives. Based on those techniques, we have designed methods for prompt detection of a queenless hive, and for prediction of a swarm being about to occur. Our highly encouraging results suggest that our Recurrent and Convolutional Neural Network-based solutions would allow reduction of the need and burden for regular, frequent physical colony inspections, a practice that disrupts the social life of the colony.

Actions (Repository Editors)

Item Control Page Item Control Page